class: center, middle, inverse, title-slide .title[ # Applying DDM on a Relational Matching task ] .author[ ### Adriana Felisa Chávez De la Peña ] .institute[ ### COGS 237; Spring 2022 ] .date[ ### April 06, 2022 ] --- class: middle, center # Week 1 Presentation --- ## Human Connectome Project (HCP) -- Collection of behavioral and neuroimaging measures for a wide range of processes. -- **Goals:** -- - Identify patterns of structural and functional connectivity in the healthy adult human brain. -- - Data available for biomedical research. --- ## Relational-Matching Task (ORIGIN) **Smith et al., (2007). A study about the rostrolateral prefrontal cortex (RLPFC)** -- - Loosely defined as the lateral portion of Brodmann area (BA) 10. -- - Associated with multiple cognitive functions across many domains. -- - **Goal:** - To develop a method for the functional + anatomical localization of RLPFC. --- ## The task  --- ## The task (Original) - Single 13-min session -- - 24 alternate blocks (12 per condition) -- - 8 trials per block (Relational) - 10 trials per block (Control) -- - Instructions flashed at the beginning of every block (1s) - "Match Change" - "Match Shape" - "Match Texture" -- - 6 textures `\(\times\)` 6 shapes. -- - Stimuli on display until - 3500 ms (Relational) - 2800 ms (Control) - Response made --- ## The task (Replica on HCP) - `\(<\)` 3-min sessions -- - 3 relational blocks - 4 trials per block - 3 control matching blocks - 5 trials per block --- ## Behavioral data -- - 82 participants (pilot for HCP). -- - 27 trials per participant. -- ```r datos <- read.csv("../00_RawMaterial/datos.csv") head(datos,10) ``` ``` ## trial subID cond dim resp corr acc rt ## 1 1 100307 relational texture 0 0 1 2548 ## 2 2 100307 relational texture 1 1 1 2528 ## 3 3 100307 relational shape 0 1 0 1690 ## 4 4 100307 relational texture 0 0 1 2438 ## 5 5 100307 control shape 1 1 1 1836 ## 6 6 100307 control texture 1 1 1 2372 ## 7 7 100307 control shape 0 0 1 2672 ## 8 8 100307 control texture 0 0 1 1724 ## 9 9 100307 control texture 1 1 1 1361 ## 10 10 100307 relational shape 0 0 1 1620 ``` --- ## Behavioral data: Response Time per Condition <!-- --> --- ## Behavioral data: Accuracy ```r accuracy <- tapply(datos$acc, datos$subID, mean) hist(accuracy, breaks = seq(0.4,1,0.05), col="cyan4", xlim=c(0,1)) ``` <!-- --> --- ## Behavioral data: Accuracy per Condition <!-- --> --- ## Drift Diffusion Model Sequential-sampling model for "rapid" binary decision making. .pull-left[ ] -- .pull-right[ **Core Assumptions** - Information is accumulated continuously over time until a response criteria is met. - RT = Decision Time + Non-decision time ] --- ## Drift Diffusion Model Sequential-sampling model for "rapid" binary decision making. .pull-left[ <br>  (Johnson et al., 2017) ] .pull-right[ **Core Assumptions** - Information is accumulated continuously over time until a response criteria is met. - RT = Decision Time + Non-decision time ] --- ## Drift Diffusion Model Sequential-sampling model for "rapid" binary decision making. .pull-left[ <br>  (Johnson et al., 2017) ] .pull-right[ **Main parameters:** - `\(\alpha\)` (Decision criteria) - Speed-accuracy tradeoff - `\(\beta\)` (Bias) - `\(z = \beta\alpha\)` (Initial point) - `\(\tau\)` (Non-decision time) - `\(\delta\)` (Drift rate) - `\(x_t \sim \mbox{N}(\delta,\sigma^2\approx0.1)\)` - Difficulty - `\(\delta \sim \mbox{N}(0,\eta)\)` ] --- ## Drift Diffusion Model Sequential-sampling model for "rapid" binary decision making. .pull-left[ <br>  (Johnson et al., 2017) ] .pull-right[ <br> <br> `$$Y = (C,T)$$` `$$Y_t \sim \mbox{Wiener}(\alpha,\beta,\tau,\delta)$$` ] --- ## Drift Diffusion Model Sequential-sampling model for "rapid" binary decision making. .pull-left[ <br>  (Johnson et al., 2017) ] .pull-right[ - `\(i^{th}\)` participant. - `\(j^{th}\)` condition. - `\(k^{th}\)` stimuli. `$$Y_{ijk} = (C_{ijk},T_{ijk})$$` `$$Y_{ijk} \sim \mbox{Wiener}(\alpha_{ijk},\beta_{ijk},\tau_{ijk},\delta_{ijk})$$` ] --- ## Drift Diffusion Model Sequential-sampling model for "rapid" binary decision making. .pull-left[ <br>  (Johnson et al., 2017) ] .pull-right[ - `\(i^{th}\)` participant. - `\(j^{th}\)` condition. - `\(k^{th}\)` stimuli. `$$Y_{ijk} = (C_{ijk},T_{ijk})$$` `$$Y_{ijk} \sim \mbox{Wiener}(\alpha_{ijk},\beta_{ijk},\tau_{ijk},\delta_{ijk})$$` `$$\alpha_{ijk} \sim *$$` `$$\beta_{ijk} \sim *$$` `$$\tau_{ijk} \sim *$$` `$$\delta_{ijk} \sim *$$` ] --- ## Drift Diffusion Model -- .pull-left[  (Vandekerckhove, Tuerlinckx and Lee, 2011) ] .pull-right[ <br> <br> <br> `$$Y_{ij} \sim \mbox{Wiener}(\alpha,\beta,\tau,\delta_{ij})$$` <br> `$$\delta_{ij} \sim \mbox{Normal}(\nu_i,\eta)$$` ] --- ## Proposal: -- - Estimate DDM parameters for the relational-matching data. -- - Implement **hierarchical Bayesian models**. -- **Phase 1: Describing individual differences** -- - Identify levels of random variation -- - Different parameter distributions by condition - Different parameter distributions by dimension - Different parameter distributions by groups-of-people -- - Which parameters? `$$Y_{ijk} \sim \mbox{Wiener}(\alpha_{ijk},\beta_{ijk},\tau_{ijk},\delta_{ijk})$$` -- - Use Bayes Factors to quantify the evidence --- ## Proposal: --  `$$\delta_{pij} \sim \mbox{Normal}(\nu_{pi},\eta_p)$$` `$$\tau_{pij} \sim \mbox{Normal}(\theta_{p},\xi_{p})$$` --- ## Proposal: -- **Phase 2: Explaining the hierarchical structure** -- - Identify predictors for these parameter variations -- - Are there any carry-over effects? - Do performance change as a function of time? - Do performance change as a function of immediately previous trial? --- class: middle, center # Thank you!